Curriculum Vitae

Youngjae Cho

ML Research Scientist working on trustworthy & robust learning — alignment, robustness, and optimization geometry — and the production pipelines that put it to work.

Experience
2025.10 — present

ML Research Scientist

Pyler · alternative military service
  • Winner — NVIDIA Nemotron Hackathon, Track B (Domain-Specialized Model), 2026
  • Implemented SimPO (reference-free RLHF) on Megatron-Bridge for Nemotron-Nano-12B-VL — post-training the framework did not support
  • First-author alignment research (GAPO) — geometry-aware preference optimization for LLMs
  • Unified LLM benchmarking CLI + pluggable engine abstraction (vLLM / Hugging Face / synthetic CI fallback)
2024.03 — 2025.10

ML Research Scientist

Aiv Co.
  • Diffusion-based defect synthesis pipeline for industrial anomaly detection (background-aware disentanglement)
  • Made diffusion finetuning + inference fit a single-GPU budget with Flash-Attention + DeepSpeed (FP16)
  • End-to-end OCR serving on Triton — ONNX→TensorRT, CUDA graphs, and test-time augmentation
Education
2022.03 — 2024.02

M.S., Industrial & Systems Engineering

KAIST · advised by Il-Chul Moon

ICML 2023 (SAAL) and AAAI 2024 (APP). Active learning, sharpness-aware optimization, Bayesian prompt adaptation.

2017.03 — 2022.02

B.S., Industrial & Systems Engineering

KAIST
Skills

Research & Modeling

PyTorchTensorFlowJAXHugging Face

Optimization & Inference

TensorRTONNXFlash AttentionDeepSpeedCUDA Graph

Serving

Triton Inference ServervLLMDocker

Domains

LLM AlignmentVision-LanguageActive LearningDiffusion / Anomaly Detection